Professional Timeline

June 2025 - August 2025

AI Engineer at Atrium

  • Collaborated with Pfizer's AI team to automate key components of Statistical Analysis Plan generation using a RAG pipeline, reducing statisticians' drafting time by up to 60%
  • Architected a custom LLM-as-a-Judge system combining prompting and structured rubric evaluation to detect hallucinations and guideline deviations in generated SAPs with over 80% precision
  • Designed Prompt Evaluation Experiments by testing variations in temperature, system prompts, and input formatting resulting in a 20% improvement in generation consistency for critical SAP tasks
Achievement: 60% reduction in drafting time, 80% precision in hallucination detection
August 2024 - Present

Master's Student at University of Maryland

  • Pursuing Master's in Applied Machine Learning at University of Maryland, College Park
  • Achieved exceptional academic performance with 3.9/4.0 GPA in first semester
  • Completed coursework in Applied Machine Learning
Achievement: 3.9/4.0 GPA
July 2022 - July 2024

Software Developer (ML Domain) at Tezo

  • Built a RAG-powered chatbot integrated with SharePoint project repositories, providing employees a unified interface to query across 10k+ internal documents including project charters, design specs, and status reports
  • Engineered a semantic search pipeline with embeddings + vector DB, reducing document lookup time by 60% and improving cross-team visibility into ongoing projects
  • Extended capabilities with LLM-based summarization of lengthy project documentation, cutting manual review time for managers by 35% and enabling quicker decision-making
  • Containerized and deployed the system with Docker + CI/CD, ensuring smooth secure on-prem access control
Achievement: 60% reduction in document lookup time, 35% reduction in manual review time
July 2021 - July 2022

Junior Software Developer (ML Domain) at Tezo

  • Migrated policy and claims data to Snowflake, streamlining ETL pipelines and reducing average query times by 40% compared to the legacy warehouse
  • Trained ML models to score policies and detect fraudulent claims, improving recall of fraud cases by 15% and enabling earlier intervention to prevent financial losses
  • Built SQL and BI dashboards for stakeholders, cutting manual report preparation time by 25%, increasing adoption to 50+ active users, and improving decision-making speed across actuarial and operations teams
Achievement: 40% reduction in query times, 15% improvement in fraud detection recall